A New Yi Font Generation Using Diffusion Model

As an important carrier of the excellent culture of the Chinese nation, the Yi script has long attracted widespread attention. However, existing research mainly focuses on feature detection, recognition, and restoration of Yi characters, while issues such as preserving glyph structure and achieving stylistic transfer remain at an exploratory stage. To address these challenges, this paper proposes a Yi font generation framework based on a multi-scale conditional diffusion model. The Multi-scale Content Perception (MCP) module is designed to capture local and global spatial information in a hierarchical manner through the constructed channel-space dual-domain multi-scale attention synergistic mechanism to improve the ability to extract content information from Yi text. Meanwhile, in order to realize the cross-modal feature fusion of Yi text, the Efficient Style Insertion (ESI) module is designed, integrating a multi-head mechanism with efficient channel attention to optimize the cross-attention mechanism. Furthermore, the sliced attention strategy is adopted to alleviate the computational overhead associated with traditional cross-attention mechanisms. The experimental results on the Yi font dataset show the advantages and feasibility of the method in this paper.